Elon: "Feature complete for full self driving this year"

[Needsdecaf]

It is funny, when you go back and read what they did promise when I bought, I am not sure it is even Level 4 and maybe not Level 3 when I read with a more Tesla critical eye today.

It says the car will take trips with no action from the driver, but notably doesn’t say the driver won’t need to pay attention during those trips.

I disagree 100%. This is where the description is VERY misleading.

It literally says "all you will need to do is get in the car and tell it where to go.". It doesn't say "all you will need to do is get in the car and tell it where to go and make sure it doesn't crash" or "be ready to take over if it needs you to" or take over when the car runs out of the area it's built for. No, it specifically says:

"All you will need to do.....".

Now, could you argue that this is NOT really implying that you won't have to monitor, won't have to be ready to take over, and all the other SAE 2/3 levels of driver involvement because it doesn't specifically state that you WON"T have to do those? Maybe. If your lawyer is expensive enough. But just like you're not supposed to use a double negative in a sentence, you shouldn't have to specifically list all the things you don't have to do in order to do something. The implication is there, clear as day, you get in your car, press a button and have no more involvement.

 

[Needsdecaf]

For all of the healthy skepticism of Tesla's optimism around the HW3 hardware achieving any higher level of autonomy, a thought did occur to me this morning.

If you want to buy a car today that has a possibility of being autonomous at some point in the future, I think Tesla is the only option. It may not be a big possibility, but given how inflexible so many other car operating systems are, Tesla is the only one that offers any sort of feasible upgrade path.

Wrong, Audi's and other VAG products have hardware for full L3 autonomy which you can actually use in Europe, but you can't use here for legal reasons.

Why the 2019 Audi A8 won't get Level 3 Traffic Jam Pilot in the US

 

[willow_hiller]

Wrong, Audi's and other VAG products have hardware for full L3 autonomy which you can actually use in Europe, but you can't use here for legal reasons.

I don't doubt that some automakers are ahead of Tesla in terms of present capability, but has Audi has equipped their vehicles in such a way as to allow the existing fleet on the road today to be upgraded past L3?

 

[diplomat33]

I disagree 100%. This is where the description is VERY misleading.

It literally says "all you will need to do is get in the car and tell it where to go.". It doesn't say "all you will need to do is get in the car and tell it where to go and make sure it doesn't crash" or "be ready to take over if it needs you to" or take over when the car runs out of the area it's built for. No, it specifically says:

"All you will need to do.....".

Now, could you argue that this is NOT really implying that you won't have to monitor, won't have to be ready to take over, and all the other SAE 2/3 levels of driver involvement because it doesn't specifically state that you WON"T have to do those? Maybe. If your lawyer is expensive enough. But just like you're not supposed to use a double negative in a sentence, you shouldn't have to specifically list all the things you don't have to do in order to do something. The implication is there, clear as day, you get in your car, press a button and have no more involvement.

"all you will need to do" in English means that it will not require anything else beyond what is mentioned. So it is implying that you will not need to supervise IMO.

However, this description of FSD was very much a future promise, not necessarily a promise of what the first iteration of FSD would be. Tesla can deliver a preliminary version of FSD that requires driver supervision while they work things out and then when the safety is good enough, then they deliver the final "all you will need to do is get in a tell it where to go" FSD without driver supervision.

 

[daniel]

I'll make a prediction that Tesla will achieve some version of autonomous driving with driver supervision with HW3 but will stumble at improving the reliability enough to remove driver supervision.

"Some version of autonomous driving" with supervision is what I have now with EAP. As long as I'm following the road I'm on and the lanes are well-marked, my car can drive itself with my supervision. And I don't need to intervene very often.

Here's your problem, you listen to what Musk promises. But in the order contract that you sign it's specifically stated that all outside promises are null and void and only things that are part of the contract are promised.

And that's been pretty much Tesla's stance in all cases I am aware of. "if it's not in the manual it was not promised" (like the on-ramp to off-ramp and meet you at doorstep summon on AP1)

View attachment 499819

I don't know why you think I have some kind of "problem." I am not talking about legal contracts. I am talking about promises made by the CEO of a company, and that have been broken. Public trust is important for a company. Being able to wriggle out of a lawsuit because the contract said "Promises made to you by sales staff and company executives are not legally binding on the company," does not help engender a feeling in car buyers that they can buy the car with confidence.

Musk promised that if I paid for FSD my car would be capable of operating as a robotaxi. My car would never have been capable of operating as a robotaxi even if I'd paid for FSD. Not only that, but now, almost two years later, if I had paid for FSD, I would not yet have any features I don't already have with EAP, and it would be at least a year before I even got the hardware to allow some screen visualizations. And it's now looking as though robotaxi is at least a decade away.

The public does not like it when a company advertises features, and then puts fine print in the contract saying that it is not legally bound by its promises.

 

[diplomat33]

"Some version of autonomous driving" with supervision is what I have now with EAP. As long as I'm following the road I'm on and the lanes are well-marked, my car can drive itself with my supervision. And I don't need to intervene very often.

No. I don't consider what we have now with EAP to be "some version of autonomous driving" because EAP is a driver assist, it is not autonomous driving. It can't stop at traffic lights or stop signs. It can't make turns at intersections. It is still missing big chunks of the DDT.

 

[Az_Rael]

I disagree 100%. This is where the description is VERY misleading.

It literally says "all you will need to do is get in the car and tell it where to go.". It doesn't say "all you will need to do is get in the car and tell it where to go and make sure it doesn't crash" or "be ready to take over if it needs you to" or take over when the car runs out of the area it's built for. No, it specifically says:

"All you will need to do.....".

Now, could you argue that this is NOT really implying that you won't have to monitor, won't have to be ready to take over, and all the other SAE 2/3 levels of driver involvement because it doesn't specifically state that you WON"T have to do those? Maybe. If your lawyer is expensive enough. But just like you're not supposed to use a double negative in a sentence, you shouldn't have to specifically list all the things you don't have to do in order to do something. The implication is there, clear as day, you get in your car, press a button and have no more involvement.

I think Tesla chose the words very carefully. I personally think Tesla will never release a Level 3 system on current hardware because they would then be liable for what the car does while your eyes are off the road.

 

[daniel]

No. I don't consider what we have now with EAP to be "some version of autonomous driving" because EAP is a driver assist, it is not autonomous driving. It can't stop at traffic lights or stop signs. It can't make turns at intersections. It is still missing big chunks of the DDT.

But where I use it I seldom need to intervene. It's true that I have to do the turns, and I don't use it on curvy roads because it doesn't stay centered in the lane when I want it to, and it will not give way when another car is too close to the line, but it stops for most traffic lights since there's usually a line of cars stopping or already stopped.

My point is merely that for maybe 75% of my driving, my car is driving itself, with infrequent interventions by me. And that percentage was much higher on long road trips, though there's no such thing where I live now.

I think Tesla would have done much better for its image and customer satisfaction if it had limited itself to mentioning L5 or robotaxi as something for the distant future and never promised it for cars being sold today. Instead of selling an "FSD" package, they should have sold an "advanced driver assist" package that would introduce unspecified features as they became available. Under-promise and over-deliver gets you much better public regard than over-promise and under-deliver, even if what you actually deliver is the same.

How long do you think it will be before Tesla can manage an intersection with stop signs on NoA? It will need to reliably recognize the stop sign and stop. It will need to understand whether it's a 4-way stop or a 2-way stop. In the former case it will need to know when it's its turn to go. In the latter it will need to recognize when cross traffic has cleared and it's safe to go, but without being so overly cautious that it blocks traffic when there is time to go. And finally, it will need to recognize when another car is going out of turn and react accordingly. This is a tall order, and until it can do this, it will not have achieved City NoA. Since we're still looking at Level 2, it will not need to be perfect, but to be of any real use, driver interventions should be needed only seldom.

 

[diplomat33]

How long do you think it will be before Tesla can manage an intersection with stop signs on NoA? It will need to reliably recognize the stop sign and stop. It will need to understand whether it's a 4-way stop or a 2-way stop. In the former case it will need to know when it's its turn to go. In the latter it will need to recognize when cross traffic has cleared and it's safe to go, but without being so overly cautious that it blocks traffic when there is time to go. And finally, it will need to recognize when another car is going out of turn and react accordingly. This is a tall order, and until it can do this, it will not have achieved City NoA. Since we're still looking at Level 2, it will not need to be perfect, but to be of any real use, driver interventions should be needed only seldom.

That's what Tesla is working on now. Yeah, it's a tall order. It requires excellent localization, mapping, and vision to see exactly where the other objects are and what they plan to do and have excellent driving policy to dictate how our Teslas should respond. It will be the first real test of Tesla's FSD approach. City driving is the hardest part of autonomous driving. It's why other companies have started with city driving and pretty much ignored highway driving. But it is doable with the right hardware and software. Mobileye can already do what you talk about:

Tesla has a version of "city NOA" working now in their internal development but it's probably not good enough yet. I expect it will probably be released to the public later this year if all goes well.

 

[DanCar]

... City driving is the hardest part of autonomous driving. ...

I've spoken to a few engineers in the business and they don't agree or disagree. The thing that makes city driving easier is that it is often at low speed. Low speed doesn't kill or maim, just a fender bender. Speed kills. For the four way stop light scenario the algo will be something like:

1. Do I think it is my turn, then inch forward
2. If none of the other cars are moving then inch forward faster

 

[diplomat33]

I've spoken to a few engineers in the business and they don't agree or disagree. The thing that makes city driving easier is that it is often at low speed. Low speed doesn't kill or maim, just a fender bender. Speed kills. For the four way stop light scenario the algo will be something like:

1. Do I think it is my turn, then inch forward
2. If none of the other cars are moving then inch forward faster

Doing things at very low speed is much easier than at 85 mph.

That's an oversimplification. City driving is often more complex and more unpredictable. You can have a pedestrian that tries to jay walk in front of you. You can have a cyclists that tries to lane split. You might need to do an unprotected left turn and not be able to view the lane to the left because a large truck is blocking your view. You might have a car pull out from a parking space and temporarily block your lane where you have to maneuver into the oncoming traffic lane to get around them while checking for oncoming traffic. Or maybe a car stalls in a roundabout and you have to maneuver around. There are a ton of situations that require careful maneuvering. Also, it is not always at harmless low speeds. There are plenty of accidents that happen because a car ran a red light at 50 mph and t-boned another car in the middle of the intersection. They can be very serious accidents. Yes, highway accidents happen at higher speeds and can be deadly but highway driving on limited access divided highways will be limited to a single direction of travel. There won't be cross traffic or traffic lights. All the cars are moving in the same direction. It makes things a lot of simpler.

Example of complex city driving:

 

[emmz0r]

That's what Tesla is working on now. Yeah, it's a tall order. It requires excellent localization, mapping, and vision to see exactly where the other objects are and what they plan to do and have excellent driving policy to dictate how our Teslas should respond.

You can't rely on HD maps, or maps at all. The GPS lives it's own life.

 

[diplomat33]

You can't rely on HD maps, or maps at all. The GPS lives it's own life.

View attachment 500082

Uh?!! You don't use GPS for localization. You use sensors like cameras or lidar to find your position, more precisely than GPS, on a HD map that you know is also accurate. In fact, companies like Mobileye and Waymo have cm accuracy, far better than what the GPS can do.

 

[emmz0r]

Uh?!! You don't use GPS for localization.

English isn't my first language, you mean like language then? Anyway what I mean is that you can't rely on GPS when it comes to intersections and stuff.

 

[diplomat33]

English isn't my first language, you mean like language then? Anyway what I mean is that you can't rely on GPS when it comes to intersections and stuff.

What I am saying is that nobody uses GPS to handle intersections since it is not accurate enough. Instead, they use cameras and/or lidar to determine the car's position on an accurate HD map. Using cameras or lidar, you can determine the position much more accurately than with GPS.

 

[emmz0r]

What I am saying is that nobody uses GPS to handle intersections since it is not accurate enough. Instead, they use cameras and/or lidar to determine the car's position on an accurate HD map. Using cameras or lidar, you can determine the position much more accurately than with GPS.

Haha, I got scared for a minute there

But GPS is used to pinpoint what 3D maps to load, if the difference between what you see and what the GPS tells you is too big, then what?

 

[Daniel in SD]

I think Tesla chose the words very carefully. I personally think Tesla will never release a Level 3 system on current hardware because they would then be liable for what the car does while your eyes are off the road.

They're releasing robotaxis this year! It's going be to be tough to avoid liability...

 

[verygreen]

press a button and have no more involvement.

you forgot "in most of the cases" and "system was designed" (but not implemented) this way.

So basically you get into the car, tell whereto go and then watch the system like a hawk to identify those circumstances where the car cannot od it on its own and then you intervene

 

[verygreen]

Public trust is important for a company. Being able to wriggle out of a lawsuit because the contract said "Promises made to you by sales staff and company executives are not legally binding on the company," does not help engender a feeling in car buyers that they can buy the car with confidence

Well, historically public trust or breaking of said trust was not a problem for both Tesla and Elon so it looks like they don't care about it all that much.

They were bitten by actual contractual promises that were notdelivered so they obviously care about it a lot more.

 

[Mardak]

but without being so overly cautious that it blocks traffic when there is time to go

Looking at how Tesla has deployed Autopilot features like Automatic Lane Change, I would guess City Street driving will be closer to the overly cautious side initially. And it wouldn't be hard to classify stopping at a stop sign then missing multiple opportunities to enter the intersection before finally entering as "feature complete" as it satisfies both stopping and determining when it's safe to enter the intersection albeit rather poorly.

And like the recent improvements to Automatic Lane Change, Tesla can adjust various parameters to make things more "aggressive" while still being safe based on fleet data. E.g., ideally Tesla has learned from human intervention to force an automatic lane change to complete when it's actually safe but Autopilot wanted to return to the original lane. But technically, human intervention for the lane change or entering an intersection weren't required.

I have low expectations of Elon Musk's "feature complete … most likely without intervention" especially around comfort level. Sure, maybe feature complete FSD successfully completed a route without requiring intervention, but it unnecessarily braked or jerked the wheel or took too long. But even then, it would be quite the milestone when that capability is released.